Investors appear to be pausing their aggressive buying of AI-related names, with semiconductor stocks falling back from recent peaks. The tech-focused Nasdaq 100 slid 3.3% on Tuesday and the broader S&P 500 closed down 1.4%, after a selloff in South Korea unnerved investors about high-flying chip shares.
Micron Technology, which had seen strong gains, dropped 13.2% in that selloff.
Why this matters
AI development is accelerating, and companies are increasingly confronting the real costs of compute — the processing power rented to run models. While prices for some compute offerings are falling, running flagship frontier models from OpenAI and Anthropic remains substantially more expensive.
At the same time demand for AI technology continues to grow, creating a mixed picture where both costs and needs shape corporate investment decisions and investor expectations.
Operating-cost visibility remains limited
A KPMG survey from May found that only 26% of the 204 U.S.-based executives polled said AI operating costs were fully visible to them, according to a report from the consulting firm highlighted earlier by the Wall Street Journal.
Rahsaan Shears, AI enterprise transformation leader at KPMG, told Axios: “Costs can get quickly out of hand.” He said clients are reporting they are burning through budgeted amounts faster than expected. For example, Uber reportedly used up its 2026 budget for AI coding tools in four months and has since limited employee spending.
Companies that had been piloting AI are now moving to scale it across their organizations, which is driving up operational spending.
Pricing dynamics and supply-demand tensions
While frontier model pricing remains high, compute costs outside those top-tier models have been declining, and some argue that AI-related stock prices are moving in step with compute costs. Deutsche Bank’s Jim Reid noted that not every business needs the most expensive models — many firms simply want “a reliable workhorse — not a supercar.”
Yet Mandeep Singh, global head of technology research at Bloomberg Intelligence, says demand for AI compute still outstrips available supply by at least five- to tenfold. He adds that falling compute prices could affect valuations for some memory-chip stocks like Micron, but have limited impact on hyperscaler valuations or broader investor expectations; in fact, lower compute costs could benefit hyperscalers that spend heavily on capacity.
Singh concluded that he does not see evidence the fundamental market drivers have changed materially.
Bottom line
The recent pullback in tech and semiconductor shares reflects a reality check in the AI buildout: companies are recognizing real operating costs as they scale, while compute pricing is evolving unevenly across segments. Madison Mills contributed reporting.
